DeepSek Shares AI Code, Strengthens Commitment To Open Source

DeepSek Founder, Liang Wenfeng (photo: x @varosbr)

JAKARTA DeepSeek, an artificial intelligence (AI) startup from China, announced it will share source codes from its five model repositories with the public next week. The move confirms the company's commitment to open-source-based AI development, an approach rarely adopted by AI companies in China or in the West.

In a post on social media platform X, DeepSek called this decision a "small but sincere progress" which will be shared in full transparency. "The basic components in our online services have been documented, implemented, and tested in production," the company wrote.

DeepSek surprised the global AI industry last month after launching an open-source R1 reasoning model, whose performance is able to compete with the West AI system at lower development costs. Unlike most AI companies that choose a closed approach, DeepSek is actually aggressively promoting openness in AI development.

DeepSek founder Liang Wenfeng, in a rare interview with Chinese media in July last year, said that his company did not prioritize the commercialization of AI models. According to him, there is a soft power in open source that can build broad influences. "Seeing other people follow our innovation provides a sense of extraordinary achievement," said Liang.

"Open source innovation is more of a cultural behavior than just a business strategy. Contributing to the development of open AI gives us respect and recognition," he added.

The source code to be shared by DeepSeek next week will provide supporting infrastructure for the AI models they have previously released, strengthening the open-source framework.

DeepSek's move was announced shortly after the company introduced its latest algorithm, the Native Sparse Attention (NSA), which is designed to increase training efficiency and inference in long context-based AI models.

Since the launch of the R1 reasoning model, the number of DeepSek users has experienced a rapid spike. According to Aicpb.com data, the site that tracks the development of AI products in China, DeepSek is now the country's most popular chatbot with 22.2 million daily active users as of January 11, 2025, surpassing Douban, which has 16.95 million users.

With this open-source strategy, DeepSeek further strengthens its position as a major player in the global AI industry, as well as opening up wider collaboration opportunities with developers communities around the world.